A method, system, device and storage medium for extracting fault characteristics of rolling bearings

Abstract

The present invention discloses a method, system, device and storage medium for extracting fault features of rolling bearings, which relates to the technical field of rotating machinery fault diagnosis, and includes: respectively constructing a sparse coefficient matrix and a dictionary for the vibration signals of the rolling bearings; constructing a robust reconstruction error function based on divergence and constructing a structured sparse double regularization constraint that simultaneously considers the intra-group sparsity and inter-group sparsity in the sparse coefficient matrix; constructing a robust structured sparse representation learning model based on the robust reconstruction error function and the structured sparse double regularization constraint, and alternately optimizing and updating the sparse coefficient matrix and the dictionary to obtain an optimal dictionary and an optimal sparse coefficient matrix; reconstructing the fault impulse signal according to the optimal dictionary and the optimal sparse coefficient matrix, and extracting the spectral features of the fault impulse signal for fault diagnosis. The robustness and accuracy of fault feature extraction are significantly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of mechanical fault diagnosis, and particularly to a method, system, device and storage medium for extracting fault features of rolling bearings. Background Art

[0002] In the early stage of the occurrence of rolling bearing faults, the weak fault transient features are easily submerged by background noise, outliers and abnormal conditions, resulting in difficulties in fault feature extraction and diagnosis.

[0003] Existing sparse fault feature extraction methods cannot simultaneously take into account the robustness to outliers and structural sparsity. The traditional square norm fidelity term is sensitive to outliers and abnormal conditions, and the analysis result is not sparse enough, resulting in interference of fault features by other components, thereby increasing the false alarm rate, and there is still room for improvement in improving the peak signal-to-noise ratio and eliminating outliers and abnormal conditions.

[0004] Moreover, existing robust sparse representation learning methods do not fully consider the inherent structural characteristics of data, resulting in poor performance in extracting weak fault features. When the transient pulse is weak, the fault features are easily ignored, affecting the accuracy of fault diagnosis. Summary of the Invention

[0005] To solve the above problems, the present invention proposes a method, system, device and storage medium for extracting fault features of rolling bearings, by designing a robust reconstruction error function and a structured sparse double regularization constraint, and constructing a robust structured sparse representation learning model therefrom, thereby significantly improving the robustness and accuracy of fault feature extraction.

[0006] To achieve the above object, the present invention adopts the following technical solutions:

[0007] In a first aspect, the present invention provides a method for extracting fault features of rolling bearings, including:

[0008] Constructing a sparse coefficient matrix and a dictionary for the vibration signal of the rolling bearing respectively;

[0009] Defining a residual signal according to the vibration signal, the sparse coefficient matrix and the dictionary, and thereby determining the divergence, and constructing a robust reconstruction error function based on the divergence by minimizing the divergence in combination with the residual signal, and constructing a structured sparse double regularization constraint that simultaneously considers the intra-group sparsity and inter-group sparsity in the sparse coefficient matrix, where is an arbitrary value not equal to 0;

[0010] Construct a robust structured sparse representation learning model based on a robust reconstruction error function and structured sparse double regularization constraints, and alternately optimize and update the sparse coefficient matrix and the dictionary until the set termination condition is met, then obtain the optimal dictionary and the optimal sparse coefficient matrix;

[0011] Reconstruct the fault pulse signal according to the optimal dictionary and the optimal sparse coefficient matrix, and extract the spectral characteristics of the fault pulse signal for fault diagnosis.

[0012] As an alternative implementation, based on the probability density function of the residual signal and the estimated probability density function obtained from the estimation of , the determined divergence is:

[0013] ;

[0014] where is the residual signal, , y is the segmented vibration signal, A is the dictionary, and x is the sparse coefficient; is the parameter in the conditional probability distribution function; is the approximation of .

[0015] As an alternative implementation, minimize the divergence is:

[0016] ;

[0017] Combined with the residual signal , the constructed robust reconstruction error function is:

[0018] ;

[0019] where is the estimation of the conditional probability distribution function parameter ; K is the number of segments of the vibration signal; is the signal matrix after segmenting the vibration signal; is the signal matrix in the i th signal segment the jth data; is the residual of the th signal segment; is the probability density function; is the intermediate parameter used to simplify the formula; is the noise standard deviation; A is the dictionary; is the sparse coefficient matrix the i column of the j-th data; is the number of points in each signal segment; is the F-norm; is the 2-norm.

[0020] As an alternative implementation, the structured sparse double regularization constraint is:

[0021] ; ;

[0022] where is the regularization constraint for controlling the sparsity within the group, is the regularization constraint for controlling the sparsity between groups; is the sparse coefficient matrix of the i column; K is the number of segments of the vibration signal; represents the p power of the norm, is the norm.

[0023] As an alternative implementation, the robust structured sparse representation learning model is:

[0024] ;

[0025] where is the residual matrix of the i column; is the i-th column of the dictionary A; is the sparse coefficient matrix of the and are the sparsity adjustment parameters.

[0026] As an alternative implementation, the dictionary update process includes:

[0027] ;

[0028] The element of the i-th row and j-th column of the diagonal weight matrix is expressed as:

[0029] ;

[0030] where is the j-th column atom of the dictionary A; K is the number of segments of the vibration signal; is the residual matrix the i column; is the sparse coefficient matrix in the i column the j-th data; is an intermediate parameter; is the noise variance.

[0031] As an alternative implementation, the sparse coefficient matrix update process includes:

[0032] ;

[0033] ; ; ;

[0034] wherein, is the sparse coefficient matrix in the i column the j-th data; is the weighted residual projection; and are both regularization threshold parameters; sign is the sign function; is the i-th column of the diagonal weight matrix ; is the T-th element of the j-th column of the dictionary A; is the number of points in each signal segment; is the residual matrix in the i column; is the j-th column atom of the dictionary A; and are sparsity adjustment parameters; is the noise standard deviation.

[0035] In a second aspect, the present invention provides a rolling bearing fault feature extraction system, including:

[0036] A preprocessing module configured to construct a sparse coefficient matrix and a dictionary for the vibration signals of the rolling bearing respectively;

[0037] A target function construction module configured to define a residual signal based on the vibration signal, the sparse coefficient matrix, and the dictionary, thereby determining the divergence, and constructing a robust reconstruction error function based on the divergence by minimizing the divergence in combination with the residual signal, and constructing a structured sparse double regularization constraint that simultaneously considers the intra-group sparsity and inter-group sparsity in the sparse coefficient matrix, where is an arbitrary value not equal to 0;

[0038] An optimization module, configured to construct a robust structured sparse representation learning model based on a robust reconstruction error function and a structured sparse double regularization constraint, and alternately optimize and update the sparse coefficient matrix and the dictionary until the set termination condition is met, so as to obtain an optimal dictionary and an optimal sparse coefficient matrix;

[0039] A feature extraction module, configured to reconstruct a fault pulse signal according to the optimal dictionary and the optimal sparse coefficient matrix, and extract the spectral features of the fault pulse signal for fault diagnosis.

[0040] In a third aspect, the present invention provides an electronic device, including a memory, a processor, and computer instructions stored on the memory and running on the processor. When the computer instructions are run by the processor, the method described in the first aspect is completed.

[0041] In a fourth aspect, the present invention provides a computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, the method described in the first aspect is completed.

[0042] In a fifth aspect, the present invention provides a computer program product, including a computer program. When the computer program is executed by a processor, the method described in the first aspect is implemented.

[0043] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0044] The present invention proposes a method, system, device and storage medium for extracting fault features of rolling bearings, and designs a robust reconstruction error function based on divergence, which effectively reduces the influence of outliers and abnormal conditions in the signal and improves the accuracy of fault feature extraction.

[0045] The present invention proposes a method, system, device and storage medium for extracting fault features of rolling bearings, and designs a structured sparse double regularization constraint, which takes into account both intra-group sparsity and inter-group sparsity and has stronger sparse representation ability.

[0046] The present invention proposes a method, system, device and storage medium for extracting fault features of rolling bearings, combines a robust reconstruction error function and a structured sparse double regularization constraint, constructs a robust structured sparse representation learning model, and has excellent performance in suppressing noise and outliers, enhancing sparsity and improving peak signal-to-noise ratio, and can effectively identify weak fault features in the signal.

[0047] Advantages of additional aspects of the present invention will be partly given in the following description, partly will become obvious from the following description, or will be understood through the practice of the present invention. Description of the Drawings

[0048] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only the embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained according to the provided drawings.

[0049] Figure 1 Flowchart of the rolling bearing fault feature extraction method provided in Embodiment 1 of the present invention. Detailed implementation manners

[0050] The following further describes the present invention in conjunction with the accompanying drawings and embodiments.

[0051] It should be noted that the following detailed descriptions are all exemplary and are intended to provide further explanations of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs.

[0052] It should be noted that the terms used herein are only for describing specific implementation manners and are not intended to limit the exemplary embodiments of the present invention. As used herein, unless otherwise clearly specified in the context, the singular form is also intended to include the plural form. In addition, it should be understood that the terms "include" and "comprise" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those clearly listed steps or units, but may include other steps or units not clearly listed or inherent to these processes, methods, products, or devices.

[0053] In the case of no conflict, the embodiments in the present invention and the features in the embodiments can be combined with each other.

[0054] Embodiment 1

[0055] This embodiment provides a rolling bearing fault feature extraction method. By designing a robust reconstruction error function and a structured sparse double regularization constraint, a robust structured sparse representation learning model is constructed, thereby significantly improving the robustness and accuracy of fault feature extraction. It includes:

[0056] Construct a sparse coefficient matrix and a dictionary for the vibration signal of the rolling bearing respectively;

[0057] Define a residual signal according to the vibration signal, the sparse coefficient matrix and the dictionary, and thereby determine divergence, and construct a based on the minimum divergence combined with the residual signal The robust reconstruction error function of divergence, and the construction of a structured sparse double regularization constraint that simultaneously considers intra-group sparsity and inter-group sparsity in the sparse coefficient matrix, where is an arbitrary value not equal to 0;

[0058] Based on the robust reconstruction error function and the structured sparse double regularization constraint, a robust structured sparse representation learning model is constructed to alternately optimize and update the sparse coefficient matrix and the dictionary until the set termination condition is met, and then the optimal dictionary and the optimal sparse coefficient matrix are obtained;

[0059] According to the optimal dictionary and the optimal sparse coefficient matrix, the fault impulse signal is reconstructed, and the spectral characteristics of the fault impulse signal are extracted for fault diagnosis.

[0060] The following combines Figure 1 to elaborate on the method of this embodiment in detail.

[0061] In this embodiment, the vibration signal of the rolling bearing is collected, and according to the sparse representation learning framework, the one-dimensional vibration signal is segmented to construct a signal matrix;

[0062] The number of signal segments is where, is the number of points in the original one-dimensional vibration signal, is the number of points in each signal segment, represents the i th vibration signal segment.

[0063] Thus, the signal matrix is expressed as:

[0064] (1).

[0065] According to the segmented vibration signal, the signal matrix is expressed as:

[0066] (2).

[0067] Since the representation ability of the analysis dictionary is limited and it cannot represent the signal adaptively, dictionary learning is adopted in this embodiment. The design of the initial dictionary has a great impact on the accuracy of fault feature extraction. The initial dictionary must be overcomplete, that is, the number of columns of the dictionary should be greater than the number of rows. In order to obtain a satisfactory representation result and meet the requirements of the robust structured sparse representation learning algorithm, the initial dictionary is constructed according to the characteristics of the fault transient signal. The fault impulse of the bearing is transient, periodic, and has time-shift invariance.

[0068] Therefore, dictionary atoms are designed by the time-domain synchronous averaging method and the cyclic shift operator. The time-domain synchronous averaging method can effectively eliminate the interference of random noise and non-periodic components while retaining the periodic components in the signal. Then, the processed vibration signal is segmented by the cyclic shift operator to construct an atom matrix, and the obtained atom matrix consists of periodic fault pulses.

[0069] Thus, the initial dictionary is expressed as:

[0070] (3);

[0071] where N is the number of points in each segment, which is used as the number of rows of the initial dictionary here, M is the number of columns of the initial dictionary, and M > N.

[0072] Based on the signal matrix and the initial fault pulse dictionary constructed above, the initial input of the robust structured sparse representation learning model is obtained.

[0073] In this embodiment, a robust structured sparse representation learning model is designed, and the objective function of this model consists of a robust reconstruction error function and a structured sparse double regularization constraint.

[0074] First, design the robust reconstruction error function; the data fidelity terms of the Frobenius norm and the norm are widely used as the reconstruction error terms, but the traditional cost function is sensitive to outliers, and the accuracy of fault feature extraction will be affected by the outliers. Therefore, in this embodiment, a robust reconstruction error function based on divergence is designed to suppress outliers and abnormal signals; where is an arbitrary value not equal to 0, and when is 1, the divergence is the KL divergence (Kulback-Leibler).

[0075] According to the morphological component analysis method, the residual signal is defined as , the residual signal , the residual of the i-th signal segment is sampled from the conditional probability distribution function , the conditional probability distribution function is the true probability distribution of the residual signal, is an unknown parameter, and the estimated conditional probability distribution parameterized by is used as the approximation of, where y is the vibration signal of the segment in the signal matrix; A is the dictionary; x is the sparse coefficient in the sparse coefficient matrix; is the parameter controlling the probability distribution in the conditional probability distribution function; is the approximate value of .

[0076] Probability density function based on the residual signal and the estimated probability density function obtained by estimating of , the divergence is defined as:

[0077] (4);

[0078] is the generalized framework of the Kullback-Leibler divergence defined by the parameter . The divergence provides a form of loss function, which is suitable for enhancing the robustness to Gaussian noise pollution.

[0079] By using the log-likelihood function , the divergence is decomposed into:

[0080] (5).

[0081] where:

[0082] (6);

[0083] (7);

[0084] (8).

[0085] By minimizing the divergence , rather than the maximum likelihood estimation, the estimated value of the parameter is obtained. Since the first term in formula (5) has nothing to do with , it has no effect on the minimization of . The minimization of the divergence, that is the estimation of

[0086] (9).

[0087] In the traditional Frobenius norm and norm data fidelity terms, the residual signal is a Gaussian white noise with a mean of zero and a variance of . Under this assumption, the model is very sensitive to outliers and anomalies. The purpose of this embodiment is to minimize the influence of noise by introducing the divergence, thereby enhancing the robustness of the model.

[0088] Probability density function is as follows:

[0089] (10);

[0090] wherein, is an element of; is the element of the i-th signal segment in the signal matrix; is the element of the dictionary A; is the element of the sparse coefficient matrix; is the noise standard deviation.

[0091] Combining Equation (8) and Equation (9), the estimation of is rewritten as:

[0092] (11);

[0093] wherein, is an intermediate parameter for simplifying the formula.

[0094] Therefore, according to the estimation derivation of the parameter this embodiment defines a robust reconstruction error function (robust fidelity):

[0095] (12);

[0096] wherein, , is the standardized residual signal.

[0097] Based on the robust reconstruction error function of the divergence reduces the sensitivity of the model to outliers and abnormal situations. To be consistent with the traditional reconstruction error term, the following derivation is carried out in this embodiment.

[0098] When and at this time, is rewritten as:

[0099] (13).

[0100] Therefore, based on the residual signal , the function is written as:

[0101] (14);

[0102] wherein, is the i j-th data in the j-th signal segment is a sparse coefficient matrix in the i column of the j-th data; is the F norm; is the 2-norm.

[0103] From the perspective of probability distribution, the robust fidelity term uses the statistical distribution of outliers and anomalies to gradually eliminate the influence on the results of sparse representation learning. Therefore, the proposed fidelity term is robust to outliers and anomalies.

[0104] Secondly, a structured sparse double regularization constraint is designed. Traditional sparse representation models only consider intra-group sparsity and do not consider inter-group sparsity, that is, the norm regularization term lacks structured information. Specifically, since the fault pulse is transient, the non-zero coefficients related to the fault characteristics are sparse in each column of the sparse coefficient matrix; at the inter-group level (the rows of the sparse coefficient matrix), due to the time-invariance and periodicity of the pulse, there is a self-similar structure between groups. Therefore, the coefficients related to the fault characteristics are also sparse at the inter-group level. To encode this structured sparsity, this embodiment designs a structured regularization to penalize the unnecessary components, so that the extracted signal is sparse.

[0105] This embodiment defines the sparse coefficient matrix of norm to introduce sparsity at both the intra-group and inter-group levels, which is described as follows:

[0106] (15);

[0107] where, is the sparse coefficient matrix of the i column, represents the p power of the norm, is norm.

[0108] In this embodiment, first, the sparse coefficient matrix is initialized, and then the sparse coefficient matrix is obtained by solving Equation (27). Most of the elements of this sparse coefficient matrix are zero, and only a few non-zero values correspond to the positions of the fault pulses.

[0109] According to the definition of the norm, the structured sparse double regularization constraint is expressed as:

[0110] (16);

[0111] (17);

[0112] Among them, is the regularization constraint for sparsity within the control group, is the regularization constraint for sparsity between control groups.

[0113] Thus, in this embodiment, based on the robust fidelity and structured sparsity constraints, a robust structured sparse representation learning model is proposed, and its formula is as follows:

[0114] (18);

[0115] Among them, is the robust reconstruction error function, and are the structured sparse double regularization constraints, is the sparsity adjustment parameter; is the data to be processed.

[0116] Therefore, the robust structured sparse representation learning model is expressed as:

[0117] (19).

[0118] The above robust structured sparse representation learning model is optimized using the exact block coordinate descent (EBCD) algorithm. The matrix is expressed as a linear weighted combination, that is . Therefore, the robust structured sparse representation learning model is rewritten as:

[0119] (20);

[0120] Among them, the first term on the right side of the equation is the robust data fidelity, and the second and third terms are the structured double regularization; is the residual matrix 's i th column, , is the sparse coefficient matrix 's i-th row, is the i-th column atom of the dictionary A; is the j-th column atom of the dictionary A; is the j-th row element of the sparse coefficient matrix .

[0121] In this embodiment, according to the robust structured sparse representation learning model, an alternating optimization strategy is adopted to update the sparse coefficient matrix and the dictionary, specifically including:

[0122] (1) Dictionary update; fix the coefficient , and the optimal dictionary atom Approximate by solving the following problem:

[0123] (21).

[0124] Differentiate Equation (21) with respect to and denote it as:

[0125] (22).

[0126] Therefore, the dictionary atom is denoted as:

[0127] (23).

[0128] To simplify the expression of the dictionary element , a diagonal weight matrix is introduced in the solution, so the updated atom is denoted as:

[0129] (24).

[0130] The diagonal weight matrix is denoted as:

[0131] (25).

[0132] The elements of the diagonal weight matrix are denoted as:

[0133] (26).

[0134] It can be seen that when the difference between and is large, that is, there are outliers and abnormal situations, and they cannot be properly fitted by , the weights of the outliers and abnormal situations are almost zero, so their influence can be eliminated.

[0135] In addition, the value in Equation (26) also affects the weights. In practical applications, the value ( is the noise variance) cannot be determined in advance. In this embodiment, a threshold is set to estimate the value. Given the residual and the threshold , if is satisfied, the actual weight satisfies , and is obtained through Equation (26), where is the weight value at the threshold.

[0136] For convenience, is fixed at 0.5 to search , and thus obtain the best experimental results. The value of should be the maximum value in the residual set . To determine the value of , assume that the value of the q -th largest element in the set is the selected ( represents rounding), and the parameter is the proportion of outliers and exceptional cases. The most important property of this weight is that it can effectively assign the smallest weighting coefficient to outliers and exceptional cases. Since the weights of outliers and exceptional cases are small, their impact on the final estimate is also small, so the proposed model is robust.

[0137] (2) Sparse coefficient update:

[0138] Fix the dictionary element , and the coefficient is approximated by solving the following problem:

[0139] (27).

[0140] Differentiate Equation (27) with respect to , which is expressed as:

[0141] (28).

[0142] The solution of

[0143] is obtained from the following formula:

[0144] (30);

[0145] (31);

[0146] (32);

[0147] where is the weighted residual projection; and are both regularization threshold parameters; sign is the sign function; is the i-th column of the diagonal weight matrix ; is the T-th element of the j-th column of the dictionary A.

[0148] In summary, the dictionary and the sparse coefficient matrix are updated through the above two alternating steps. When the termination condition is satisfied, where To tolerate errors, the final optimal dictionary and the optimal sparse coefficient matrix are output.

[0149] In this embodiment, according to the optimal dictionary and the optimal sparse coefficient matrix , the impulse signal matrix is obtained, and then the one-dimensional estimated fault impulse signal is reconstructed therefrom. The estimated fault impulse signal is analyzed and processed using the squared envelope spectrum (SES) to obtain the spectral characteristics. Based on the analyzed spectral characteristics and the theoretical fault characteristic frequencies, fault diagnosis is carried out.

[0150] The method proposed in this embodiment includes four main steps: weight calculation, residual calculation, sparse coefficient update, and dictionary atom update. To illustrate the performance of the proposed method, the calculation of its complexity includes:

[0151] During K iterations, the calculation of the weight requires operations, and the calculation of the residual requires operations. In the sparse coefficient update stage, the calculations of , and require operations respectively. Therefore, the sparse coefficient update stage requires operations. In the dictionary atom update stage, and require and operations respectively, and the calculation of each atom requires operations. The complexity of updating M atoms is . Therefore, the total complexity of the proposed algorithm is . The computational complexity is not high, and less computing resources are used, which can improve efficiency.

[0152] In this embodiment, the experimental equipment includes a motor, an acceleration sensor, a thermocouple, and a bearing. During the experiment, the motor speed is maintained at 2000 RPM. The fault characteristic frequency of the outer raceway is calculated to be 236.4 Hz.

[0153] The proposed robust structured sparse representation learning model includes trade-off parameters and , ratio parameters And the number of dictionary atoms M. To study the influence of these parameters in the bearing fault feature extraction experiment, in this embodiment, different and parameter combinations are used to compare the peak signal-to-noise ratio. The results show that the ranges of the trade-off parameters and are respectively , . Subsequently, the trade-off parameters , are fixed, and the influence of the proportional parameter is studied. Since represents the proportion of outliers and abnormal conditions in the residual , its range is [0, 1], and the specific value cannot be determined in advance. To obtain the best experimental performance, a parameter search (step size is set to 0.01) is performed on . The results show that when are set to 0.19 and 0.30 respectively, the proposed method has the best effect. The number of dictionary atoms M needs to be selected by weighing between the computational cost and the required optimal peak signal-to-noise ratio.

[0154] In this embodiment, the proposed robust structured sparse representation learning method is used for bearing fault feature extraction. To obtain better performance, the number of initial dictionary atoms is set to M = 40. According to the parameter search results, the trade-off regularization parameters are set to , . The parameter for rolling bearing fault feature extraction is set to 0.19 to further improve the performance. The experimental results show that the proposed robust structured sparse representation learning method can effectively extract fault pulses from time-domain signals. In the frequency domain, , , , , and the spectral peaks at are clearly visible. The clear spectral peaks at the fault feature frequencies can provide effective indicators for judging the outer ring fault of the bearing. The invention shows excellent performance in suppressing noise and outliers, and at the same time improves the sparsity, providing an effective solution for the fault diagnosis of rolling bearings.

[0155] Embodiment 2

[0156] This embodiment provides a rolling bearing fault feature extraction system, including:

[0157] A preprocessing module configured to construct a sparse coefficient matrix and a dictionary for the vibration signal of the rolling bearing respectively;

[0158] The objective function construction module is configured to define a residual signal based on a vibration signal, a sparse coefficient matrix, and a dictionary, thereby determining the divergence, and constructing a robust reconstruction error function based on the divergence combined with the residual signal by minimizing the divergence, and constructing a structured sparse double regularization constraint that simultaneously considers intra-group sparsity and inter-group sparsity in the sparse coefficient matrix, where is an arbitrary value not equal to 0;

[0159] The optimization module is configured to construct a robust structured sparse representation learning model based on the robust reconstruction error function and the structured sparse double regularization constraint, thereby alternately optimizing and updating the sparse coefficient matrix and the dictionary until the set termination condition is satisfied, and then obtaining the optimal dictionary and the optimal sparse coefficient matrix;

[0160] The feature extraction module is configured to reconstruct a fault pulse signal according to the optimal dictionary and the optimal sparse coefficient matrix, and extract the spectral features of the fault pulse signal for fault diagnosis.

[0161] It should be noted here that the above modules correspond to the steps described in Embodiment 1, and the examples and application scenarios implemented by the above modules and the corresponding steps are the same, but are not limited to the content disclosed in the above Embodiment 1. It should be noted that the above modules can be executed in a computer system such as a set of computer executable instructions as part of the system.

[0162] In more embodiments, there is also provided:

[0163] An electronic device includes a memory, a processor, and computer instructions stored on the memory and running on the processor. When the computer instructions are run by the processor, the method described in Embodiment 1 is completed. For the sake of brevity, it will not be elaborated here.

[0164] It should be understood that in this embodiment, the processor may be a central processing unit CPU, and the processor may also be other general-purpose processors, digital signal processors DSP, application-specific integrated circuits ASIC, off-the-shelf programmable gate arrays FPGA, or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0165] The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A part of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0166] A computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the method described in Embodiment 1.

[0167] The method in Embodiment 1 can be directly implemented by a hardware processor, or by a combination of hardware and software modules in the processor. The software module can be located in a mature storage medium in the art such as random access memory, flash memory, read-only memory, programmable read-only memory, or electrically erasable programmable memory, register, etc. This storage medium is located in the memory, and the processor reads the information in the memory and combines its hardware to complete the steps of the above method. To avoid repetition, it will not be described in detail here.

[0168] A computer program product includes a computer program, which, when executed by a processor, implements the method described in Embodiment 1.

[0169] The present invention also provides at least one computer program product tangibly stored on a non-transitory computer-readable storage medium. The computer program product includes computer-executable instructions, such as instructions included in program modules, which are executed in a device on a target real or virtual processor to perform the process / method as described above. Generally, program modules include routines, programs, libraries, objects, classes, components, data structures, etc. that perform specific tasks or implement specific abstract data types. In various embodiments, the functions of program modules can be combined or divided as needed. The machine-executable instructions for program modules can be executed locally or within a distributed device. In a distributed device, program modules can be located in local and remote storage media.

[0170] The computer program code for implementing the method of the present invention can be written in one or more programming languages. These computer program codes can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing devices, so that when the program codes are executed by the computer or other programmable data processing devices, the functions / operations specified in the flowchart and / or block diagram are implemented. The program codes can be executed entirely on the computer, partially on the computer, as an independent software package, partially on the computer and partially on a remote computer, or entirely on a remote computer or server.

[0171] In the context of the present invention, the computer program code or related data can be carried by any suitable carrier so that the device, apparatus, or processor can perform the various processes and operations described above. Examples of carriers include signals, computer-readable media, etc. Examples of signals can include electrical, optical, radio, sound, or other forms of propagated signals, such as carrier waves, infrared signals, etc.

[0172] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in connection with this embodiment can be implemented by electronic hardware or a combination of computer software and electronic hardware. Whether these functions are executed in hardware or software depends on the specific application and design constraints of the technical solution. Skilled artisans may use different methods for each specific application to implement the described functions, but such implementation should not be considered to exceed the scope of this application.

[0173] Although the specific implementation manners of the present invention have been described above in conjunction with the accompanying drawings, it is not a limitation on the protection scope of the present invention. Those skilled in the art should understand that, based on the technical solution of the present invention, various modifications or deformations that can be made by those skilled in the art without creative efforts are still within the protection scope of the present invention.

Claims

1. A rolling bearing fault feature extraction method, characterized in that: include: Construct sparse coefficient matrix and dictionary for the vibration signal of rolling bearing respectively; The residual signal is defined based on the vibration signal, the sparse coefficient matrix and the dictionary to determine Divergence, by minimizing Divergence combined with residual signal is constructed based on The robust reconstruction error function of the divergence is constructed, and a structured sparse dual regularization constraint is constructed that simultaneously considers the intra-group sparsity and inter-group sparsity in the sparse coefficient matrix, where is any value not equal to 0; A robust structured sparse representation learning model is constructed based on the robust reconstruction error function and structured sparse dual regularization constraints, so as to alternately optimize and update the sparse coefficient matrix and dictionary until the set termination conditions are met, and then the optimal dictionary and the optimal sparse coefficient matrix are obtained. Reconstructing the fault pulse signal according to the optimal dictionary and the optimal sparse coefficient matrix, and extracting the frequency spectrum characteristics of the fault pulse signal for fault diagnosis; minimize The divergence is: ； Combined residual signal , the robust reconstruction error function constructed for: ； in, is the conditional probability distribution function parameter The estimation of; K is the number of segments of the vibration signal; is the signal matrix after the vibration signal is segmented; The signal matrix Middle i Signal segment The jth data of Segment the i-th signal The residual of is the probability density function; is an intermediate parameter used to simplify the formula; is the noise standard deviation; A is the dictionary; is a sparse coefficient matrix Middle i List The jth data of is the number of points in each signal segment; is the F-norm; is the 2-norm; The structured sparse dual regularization constraint is: ； ； in, is a regularization constraint controlling the sparsity within a group, is a regularization constraint that controls the sparsity between groups; is a sparse coefficient matrix No. i List; express Norm p Power, yes Norm.

2. A rolling bearing fault feature extraction method as claimed in claim 1, characterized in that: Probability density function based on residual signal and The estimated probability density function is , sure The divergence is: ； in, is the residual signal, , y is the segmented vibration signal, A is the dictionary, and x is the sparse coefficient; is the parameter in the conditional probability distribution function; for Approximate value of .

3. A rolling bearing fault feature extraction method as claimed in claim 1, characterized in that: The robust structured sparse representation learning model is: ； in, is the residual matrix No. i List; is the i-th column of dictionary A; is a sparse coefficient matrix The jth row in the middle; and is the sparsity tuning parameter.

4. A rolling bearing fault feature extraction method as claimed in claim 1, characterized in that: The dictionary update process includes: ； Diagonal weight matrix The element in the i-th column and j-th row of It is expressed as: ； in, is the j-th column atom of dictionary A; K is the number of segments of the vibration signal; is the residual matrix No. i List; is a sparse coefficient matrix Middle i List The jth data of is the intermediate parameter; is the noise variance.

5. A rolling bearing fault feature extraction method as claimed in claim 1, characterized in that: The sparse coefficient matrix update process includes: ； ； ； ； in, is a sparse coefficient matrix Middle i List The jth data of is the weighted residual projection; and are all regularization threshold parameters; sign is the sign function; is the diagonal weight matrix The i-th column of is the Tth element in the jth column of dictionary A; is the number of points in each signal segment; is the residual matrix No. i List; is the j-th column atom of dictionary A; and is the sparsity adjustment parameter; is the noise standard deviation.

6. A rolling bearing fault feature extraction system, characterized in that: A rolling bearing fault feature extraction method for implementing any one of claims 1 to 5, comprising: A preprocessing module is configured to construct a sparse coefficient matrix and a dictionary for the vibration signal of the rolling bearing respectively; The objective function building module is configured to define a residual signal based on the vibration signal, the sparse coefficient matrix and the dictionary to determine Divergence, by minimizing Divergence combined with residual signal is constructed based on The robust reconstruction error function of the divergence is constructed, and a structured sparse dual regularization constraint is constructed that simultaneously considers the intra-group sparsity and inter-group sparsity in the sparse coefficient matrix, where is any value not equal to 0; An optimization module is configured to construct a robust structured sparse representation learning model based on a robust reconstruction error function and a structured sparse dual regularization constraint, so as to alternately optimize and update the sparse coefficient matrix and the dictionary until a set termination condition is met, thereby obtaining an optimal dictionary and an optimal sparse coefficient matrix; The feature extraction module is configured to reconstruct the fault pulse signal according to the optimal dictionary and the optimal sparse coefficient matrix, and extract the frequency spectrum features of the fault pulse signal for fault diagnosis.

7. An electronic device, characterized in that: The method comprises a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method according to any one of claims 1 to 5 is completed.

8. A computer-readable storage medium, characterized in that: Used to store computer instructions, which, when executed by a processor, complete the method described in any one of claims 1 to 5.

Images